Last week was one of my favorite weeks of the year. No, not my birthday week. It was O’Reilly’s Velocity Conference in Santa Clara. I’ve attended Velocity five of the last six years, and it is always a great time. I spent the week reconnecting with former co-workers and friends, putting faces with Twitter avatars, and totally geeking out about web performance.
It is always a struggle for me to choose which sessions to attend. It seems like there are two or three sessions I find appealing during every time slot, not to mention the hallway track conversations and checking out the new technologies in the exhibit hall.
Below are some talks that piqued my interest:
Initial congestion window size
On Tuesday, I attended Tobias Baldauf’s tutorial on HTTP/2 and Hero Images. I was intrigued by his formula to calculate the best initial congestion window and the idea that the initial congestion window size in HTTP/1.1 may result in less than optimal performance in HTTP/2. The more I thought about this, the more confused I got. Currently there is an experimental specification RFC 6298 to increase TCP’s initial window to 10, but Tobias was talking about increasing it to 25. Is there another proposal forthcoming to propose changing the max initial congestion window from 10 to 25?
Today we are supporting HTTP/1.1 and HTTP/2. What mechanisms are in place to support optimal settings for each of these protocols? Or, are we faced with having sub-optimal performance for one protocol while we optimize for the other?
Lots of conversations and talks about the impact of browser extensions, in particular Ad blockers.
- Bruce Lawson, Deputy CTO at Opera, gave an excellent keynote, Making bad ads sad. Rad!, filled with statistics and humor on the state of advertising on the web.
- Instart Logic’s Karan Kumar presentation on Measuring Ad Blocker's Impact on Performance provided a slightly different take on ad blockers, and reported results from a study on how ad blockers don’t always improve performance. For sites that are ad heavy, ad blockers can improve performance – but when the site has limited ads, or none, performance can actually worsen.
- I’m disappointed I missed the talk from Yoav Weiss on Taking Back Control over third-party content. I would have liked to hear how that compared to the other two talks. I will have to watch the session replay.
Tammy Everts and Pat Meenan presented Using machine learning to determine drivers of bounce and conversion. Their talk raised a lot of questions that need further exploration, as their results differ from previous studies and many people’s gut reactions.
One question I had was, why does the number of scripts impact conversions? I found myself wondering if pages with fewer scripts actually had heavier scripts, and if the true issue was weight of the scripts, not the number.
It was disappointing to hear that metrics we focus on in terms of perceived user experience turned out to be relatively meaningless in their impact on bounce rates and conversions. A large part of me wants this information to be wrong: perception of page load time has to have an impact. I may just bury my head in the sand and pretend I didn’t hear that so I can keep talking about perception and user experience – at least until I see more studies corroborating this.
Estelle Weyl is trying to make the web accessible as well as fast. In her Accessibility as Performance talk, she illustrated how making a site accessible does not mean you have to sacrifice performance. On the contrary: Estelle demonstrated how using semantic HTML can improve both accessibility and performance
There are many talks I wasn’t able to attend, such as
- The top 10 performance defects that are costing you millions
- H2 in the real world
- Understanding the performance penalty: HTTP vs HTTPS
- Facebook web performance: Using data from the field
If you missed Velocity, don’t worry – there are many slides and videos available at the conference site, and Andreas Grabner has provided the notes he took during keynotes and sessions on his blog. I will spend a good portion of the coming weeks checking out presentations and videos of the sessions I missed because, alas, I still don’t own a TARDIS.
Until next year...